Imputation of income on the LS

Chris White, Office for National Statistics

[Project number 20067]

Research using the ONS Longitudinal Study has indicated that socio-economic inequalities in health have persisted since 1972. However, due to the fact that income is not collected at census we have been unable to look at the relationship between income and mortality, and instead, have used measures such as social class or housing tenure.

Research in other countries suggests a strong relationship between income and mortality, and in some cases between income inequality and mortality. The relationship differs between westernised countries and is of interest. Imputing income on the LS would go some way to estimating whether or not we are in a similar situation to the US, or to Canada for example.

Imputing income would also enable researchers to determine if area variation in the relationship between income, income inequality and mortality is wholly explained by the distribution of income among individuals. This will be made possible due to the fact that ward level imputed income will be available from March 2003.

Income estimates have already been added to the Sample of Anonymised Records. However this estimation, based on ward level estimates for households, could be improved given the level of data available at the individual level on the LS.

There are a number of other uses for an income estimate, including simply trying to see if use of such an estimate better explains variations in mortality than does social class. An income estimate could expand the number of users of the LS.

This project aims to create an income estimate to add to the LS dataset by:

1. Identifying an optimal donor dataset (s)
2. Building a predictive equation for income based on variables available in the LS on this donor dataset.
3. Applying this equation to the LS.

This equation would produce an estimate of income, an estimate of the random error and an estimate of the variation in this random error. Thus achieving an income estimate and an estimate of its reliability as an indicator.

Top of page